Spatial distribution pattern of tau depositions in Alzheimer’s disease using data-driven approach of flortaucipir PET

Ziel/Aim One of the main hallmarks of Alzheimer’s disease (AD) is the neurofibrillary tangles (NFTs), which are composed of tau protein. Tau PET imaging became a useful tool for the early diagnosis of AD and the detection of the disease progression. However, the interpretation of spatial patterns of tau PET associated with pathologic progression is still limited. Here, we introduce an unsupervised learning, data-driven approach to discover spatial features of tau deposition from PET imaging.

Methodik/Methods 1080 (130 AD, 588 CN, 339 MCI, 23 unidentified) pairs of AV-1451 tau PET images and MRI T1 images were obtained from the ADNI database. All PET images were spatially normalized using statistical parametric mapping (SPM8). Variational autoencoder was built to derive latent features of spatial pattern of tau depositions. The extracted latent features were used as an input for the agglomerative clustering. t-distributed stochastic neighborhood embedding (t-SNE) was utilized to visualize clustering results. The extended diagnosis, including the converters from MCI to AD, was compared with clustering result. The cluster difference was evaluated using one-way analysis of variance (ANOVA) with sex and APOE status and fisher’s exact test with age and MMSE score. Turkey’s posthoc pair-wise test for multiple comparisons was followed. The average tau standardized uptake value ratio (SUVr) in various regions using cerebellum grey matter as a reference region was estimated.

Ergebnisse/Results t-SNE plot and the contingency matrix showed the significance of the clustering result. Four clusters according to the spatial pattern of tau PET were identified. AD patients and converters were significantly more in a specific cluster, ‘cluster 3ʹ. Cluster 3 had higher average tau SUVr overall, especially in the cingulate and frontal region. The cingulate SUVr in cluster 3 had higher variation among subjects than other clusters.

Schlussfolgerungen/Conclusions The suggested work has potential to support the diagnosis and prognosis of Alzheimer’s Disease.

Publication History

Article published online:
08 April 2021